Re: [PATCH] net: phy: phy_ethtool_ksettings_set: Allow any supported speed

[Andrew Lunn]

On Mon, Dec 02, 2024 at 04:09:43PM +0500, Nikita Yushchenko wrote:
> > > Right now, 'ethtool -s tsn0 master-slave forced-slave' causes a call to
> > > driver's ethtool set_link_ksettings method. Which does error out for me
> > > because at the call time, speed field is 2500.
> > 
> > Are you saying that the PHY starts in fixed-speed 2.5G mode?
> > 
> > What does ethtool tsn0 say after boot and the link has come up but
> > before any ethtool settings are changed?
> 
> On a freshly booted board, with /etc/systemd/network temporary moved away.
> 
> (there are two identical boards, connected to each other)
> 
> root@vc4-033:~# ip l show dev tsn0
> 19: tsn0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN mode DEFAULT group default qlen 1000
>     link/ether 3a:e3:5c:56:ba:bd brd ff:ff:ff:ff:ff:ff
> 
> root@vc4-033:~# ethtool tsn0
> Settings for tsn0:
>         Supported ports: [ MII ]
>         Supported link modes:   2500baseT/Full

If it is a T1 PHY, we want it reporting 25000BaseT1/Full here.  Having
T1 then probably allows us to unlock forced master/slave without
autoneg, and setting speeds above 1G without autoneg.

Given this is an out of tree driver, i can understand why it does not,
it means patching a number of in tree files.

https://www.marvell.com/products/automotive/88q4364.html

says it can actually do 2.5G/5G/10GBASE-T1 as defined by the IEEE
802.3ch standard.

I would be reluctant to make changes to phylib without a kernel
quality PHY driver queued for merging. So you might want to spend some
time cleaning up the code. FYI: I've not looked at 802.3ch, but if
that defines registers, i would expect the driver patches to actually
be split into helpers for standard defined registers any 3ch driver
can use, and a PHY driver gluing those together and accessing marvell
specific registers.

	 Andrew